Semi-supervised image analysis method, device, computer equipment and storage medium
By distinguishing between causal and non-causal features and optimizing the objective loss function, the problem of unclear distinction between causal and non-causal features in existing semi-supervised learning methods is solved, improving the model's generalization ability and accuracy, especially in medical image classification performance when labeled samples are scarce.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN INST OF ADVANCED TECH CHINESE ACAD OF SCI
- Filing Date
- 2024-08-20
- Publication Date
- 2026-05-15
AI Technical Summary
Existing semi-supervised learning methods in medical image analysis often fail to distinguish between causal and non-causal features, leading to spurious consistency alignment learning and affecting the model's generalization ability and accuracy.
By distinguishing between causal and non-causal features, an attention mechanism is used to generate a causal score mask. The teacher branch parameters are updated by combining exponential moving average. Mutual information constraints and causal consistency alignment loss are introduced to optimize the objective loss function for training the student model.
It improves the model's generalization ability and accuracy, especially in cases where labeled samples are scarce, significantly enhancing the performance of medical image classification.
Smart Images
Figure CN118941928B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of semi-supervised image analysis technology, and in particular to a semi-supervised image analysis method, apparatus, computer equipment, and storage medium. Background Technology
[0002] With the rapid development of deep learning and computer vision, computer-aided diagnostic medical image analysis has achieved unprecedented performance. Existing methods typically focus on supervised learning models, which require large-scale labeled sample datasets. However, unlike natural images, medical image annotation must be performed by experienced clinicians, making the process time-consuming, expensive, and challenging. Therefore, leveraging a large number of unlabeled samples to improve the accuracy of machine learning models, especially when labeled samples are scarce, is of significant research value in the field of medical image analysis. Semi-supervised learning has important theoretical and practical significance in improving model performance using limited labeled data and a large amount of unlabeled data.
[0003] In semi-supervised learning methods, pseudo-labeling is often constrained by threshold selection strategies. It involves labeling unlabeled samples (pseudo-labels) using algorithms, then using these pseudo-labels as new labeled samples for model retraining. These methods typically face a trade-off in threshold selection. A strict threshold results in better quality pseudo-labels, but fewer new samples are used for retraining each time, and vice versa. On the other hand, consistency regularization techniques are also commonly used in semi-supervised learning, such as Temporal Ensembling models and Mean-Teacher models. They aim to maintain consistency in similar logical distributions (e.g., predictions) across different views of the same sample, thereby enhancing the semantic representation of the target. However, since these models do not explicitly distinguish between causal and non-causal features in different perturbations, non-causal but distinguishable features may replace inherent semantic information when considering invariant constraints, leading to spurious consistency. Therefore, applying traditional consistency constraints to indiscriminately align sample features or prediction distributions by utilizing unlabeled data may lead to suboptimal results. Considering the weaknesses of existing consistency regularization methods, a re-evaluation of these methods is necessary. Traditional methods indiscriminately enforce consistency across all parts of batches in both teacher and student models. Such extensive enforcement often includes a great deal of invalid and obfuscated information, which provides little to no valuable learning information. These obfuscated factors can not only distract from the learning process but also dilute the content of effective semantic information, thereby undermining the potential advantages of consistency-based regularization methods. Summary of the Invention
[0004] Therefore, it is necessary to address the technical problem that the existing semi-supervised image analysis technology has poor performance, and propose a semi-supervised image analysis method, device, computer equipment and storage medium.
[0005] Firstly, a semi-supervised image analysis method is provided, the method comprising:
[0006] Obtain sample images from the training set;
[0007] Based on the sample images and the initial student model, predictions are made to obtain the first prediction distribution corresponding to the non-causal features and the second prediction distribution corresponding to the first causal features.
[0008] Based on the sample images and the initial teacher model, a third prediction distribution corresponding to the second causal feature is obtained;
[0009] Based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature, the target loss function is determined;
[0010] Based on the target loss function, the network parameters in the initial student model are updated, and the updated initial student model is used as the trained student model.
[0011] Secondly, a semi-supervised image analysis apparatus is provided, the apparatus comprising:
[0012] The acquisition module is used to acquire sample images from the training set;
[0013] The first prediction module is used to make predictions based on the sample images and the initial student model to obtain a first prediction distribution corresponding to non-causal features and a second prediction distribution corresponding to the first causal features.
[0014] The second prediction module is used to make predictions based on the sample images and the initial teacher model to obtain the third prediction distribution corresponding to the second causal feature.
[0015] The determination module is used to determine the target loss function based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature;
[0016] The training module is used to update the network parameters in the initial student model based on the target loss function, and use the updated initial student model as the trained student model.
[0017] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the semi-supervised image analysis method described above.
[0018] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described semi-supervised image analysis method.
[0019] The semi-supervised image analysis method proposed in this invention acquires sample images from a training set, then performs predictions based on these sample images and an initial student model to obtain a first prediction distribution corresponding to non-causal features and a second prediction distribution corresponding to the first causal features. Next, based on the sample images and an initial teacher model, it performs predictions to obtain a third prediction distribution corresponding to the second causal features. Based on the first, second, and third prediction distributions, the first and second causal features, and the non-causal features, a target loss function is determined. Finally, based on the target loss function, the network parameters in the initial student model are updated, and the updated initial student model is used as the trained student model. This method enables the initial student model to be trained using a target loss function, resulting in a trained student model with strong generalization ability, high efficiency, and high accuracy. Using this student model for image analysis improves the effectiveness of image analysis. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0021] in:
[0022] Figure 1 This is a diagram illustrating the application environment of a semi-supervised image analysis method in one embodiment.
[0023] Figure 2 A flowchart of a semi-supervised image analysis method in one embodiment;
[0024] Figure 3 This is a schematic diagram of the structure of a semi-supervised image analysis method in one embodiment;
[0025] Figure 4 This is a block diagram of a semi-supervised image analysis device in one embodiment;
[0026] Figure 5 This is a structural block diagram of a computer device in one embodiment;
[0027] Figure 6 This is a structural block diagram of a computer device in another embodiment. Detailed Implementation
[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.
[0029] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0031] The semi-supervised image analysis method provided in this invention can be applied to, for example... Figure 1In this application environment, client 110 communicates with server 120 via a network. Server 120 receives sample images from the training set through client 110. Then, server 120 performs predictions based on the sample images and an initial student model, obtaining a first prediction distribution corresponding to non-causal features and a second prediction distribution corresponding to the first causal features. Next, server 120 performs predictions based on the sample images and an initial teacher model, obtaining a third prediction distribution corresponding to the second causal features. Based on the first, second, and third prediction distributions, the first and second causal features, and the non-causal features, server 120 determines a target loss function. Finally, server 120 updates the network parameters in the initial student model based on the target loss function, using the updated initial student model as the trained student model. This allows training the initial student model using the target loss function, resulting in a trained student model with strong generalization ability, high efficiency, and high accuracy. This student model is then used for image analysis, improving the effectiveness of image analysis. Client 110 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server 120 can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.
[0032] Please see Figure 2 As shown, Figure 2 A flowchart illustrating a semi-supervised image analysis method according to an embodiment of the present invention includes the following steps:
[0033] Step S101: Obtain sample images from the training set;
[0034] The sample images may include labeled and unlabeled samples, and the sample images may be medical images, such as images of human organs.
[0035] Step S102: Based on the sample image and the initial student model, make predictions to obtain the first prediction distribution corresponding to the non-causal features and the second prediction distribution corresponding to the first causal features;
[0036] In one embodiment, a first feature is obtained based on the sample image input and the encoder in the initial student model; a first mask is obtained based on the first feature and the scoring function, and a second mask is determined based on the first mask; a first causal feature is obtained based on the first feature and the first mask, and a non-causal feature is obtained based on the first feature and the second mask; a first prediction distribution corresponding to the non-causal feature is obtained based on the first linear classifier and the non-causal feature in the initial student model; and a second prediction distribution corresponding to the first causal feature is obtained based on the second linear classifier and the first causal feature in the initial student model.
[0037] Specifically, such as Figure 3 As shown, to achieve effective causal feature learning, causal and non-causal features must be separated. This work employs an attention mechanism to distinguish between causal and non-causal components, namely, a causal indicator. Given an encoder F in the initial student model... s (.), using the scoring function f on the final convolutional features of the encoder output. score To estimate the importance scores of causal components and generate a causal score mask M. C The M C This refers to the first mask. The generated causal fraction mask has the same spatial resolution as the feature map. Similarly, masks M for irrelevant / non-causal features can be obtained. B M B This refers to the second mask:
[0038] M C =σ(f score (F s (x i ;θ s ,π s )),θ m )
[0039] M B =1-M C
[0040] Where, f score The parameter is θ m The convolutional layer is used to project the feature components at each location to obtain a score; θ s The student branch network refers to the parameters of the encoder network, π. s Indicates the sample x i The perturbation, σ(.), represents the sigmoid function used to further normalize the fractional result to the (0,1) interval. The first characteristic refers to F. s (x i ;θ s ,π sAfter defining the causal indicator, it is necessary to extract causal and bias features. Specifically, the original feature embedding (first feature) is decomposed into first causal features. Sum of deviation characteristics This deviation characteristic This refers to non-causal features, and is processed by a linear classifier. and The following classification predictions were obtained:
[0041]
[0042] Where ⊙ represents the Hadama product, Represents the first linear classifier The parameters, Represents the second linear classifier The parameters of causal features aim to obtain core information about a disease; they should have the ability to distinguish disease categories. This refers to the second prediction distribution. This refers to the first predictive distribution and the first causal feature. Non-causal characteristics
[0043] Step S103: Based on the sample image and the initial teacher model, make a prediction to obtain the third prediction distribution corresponding to the second causal feature;
[0044] In one embodiment, the step of predicting the third prediction distribution corresponding to the second causal feature based on the sample image and the initial teacher model is as follows:
[0045] Step S1031: Based on the sample image input and the encoder in the initial student model, obtain the second feature;
[0046] Step S1032: Based on the second feature and the first mask obtained by sharing the initial teacher model, obtain the second causal feature;
[0047] Step S1033: Based on the third linear classifier and the second causal feature in the initial teacher model, obtain the third prediction distribution corresponding to the second causal feature.
[0048] In this embodiment, an exponential moving average is applied to update the parameters of the teacher branch. Throughout the entire framework optimization process, only the initial student network is updated. Specifically, the teacher model parameters θ at the T-th iteration... t The update is as follows: Where α represents the weighting coefficient.
[0049] In one embodiment, the initial teacher and initial student branches share the learned causal mask M. CSimilarly, the second causal feature of the teacher branch can be obtained. and the third prediction distribution
[0050]
[0051] The encoder in the initial teacher model is F. t , This is the second causal characteristic. The third predicted distribution, Represents the third linear classifier The parameters.
[0052] Step S104: Determine the target loss function based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature;
[0053] In one embodiment, the target loss function includes a supervised loss, which is represented as follows:
[0054]
[0055] in, Represented as the second prediction distribution, l ce Represents the cross-entropy loss, y i N represents the true label. l x represents the number of labeled samples. i For sample images, To monitor losses;
[0056] Meanwhile, the bias component contains content unrelated to disease identification and cannot be used as a classification clue. Therefore, the target loss function also includes alignment loss, which is expressed as follows:
[0057]
[0058] in, For the first prediction distribution, y unif For a uniform distribution, KL represents the Kullback-Leibler entropy. For alignment loss.
[0059] In one embodiment, to make robust predictions, it is insufficient to merely emphasize causally significant features. It is equally important to ensure that these causal features are distinct and independent from irrelevant non-causal features. This separation ensures that the model does not rely on spurious correlations between the causal and non-causal components, thus achieving better generalization. The dependence between different features can be measured using mutual information (MI). The goal is to minimize mutual information. To restrict the first causal feature and bias To address the dependency between them, the target loss function also includes a mutual information function, which is expressed as follows:
[0060]
[0061] Where, q θ (.|.) denotes a variational approximation of the true distribution. This refers to the first causal characteristic. It refers to non-causal characteristics; MI refers to mutual information. This refers to the mutual information function. By introducing the MI independence constraint during supervised learning, the model can better learn representations where causal and non-causal features are clearly separated. This approach ensures that the model's decision-making process is primarily influenced by causal features, and that causal features are better able to predict the true results of medical image classification tasks.
[0062] In one embodiment, to improve the performance of consistency learning, within the framework of the invention, causal features or causal predictions of the student and teacher models are used for constraints under different reinforcement methods. Without loss of generality, two causal consistency alignment losses are defined: a first causal consistency alignment loss and a second causal consistency alignment loss.
[0063] The objective loss function further includes a first causal consistency alignment loss, which is expressed as follows:
[0064]
[0065] Where, N u The number of unlabeled samples in the training set. This refers to the third prediction distribution. This represents the first causal consistency alignment loss;
[0066] The objective loss function also includes a second causal consistency alignment loss, which is expressed as follows:
[0067]
[0068] in, The second causal consistency alignment loss describes the inter-sample consistency mechanism based on causal alignment, which constrains the consistency of causal feature outputs from different branches; R represents the relation definition function, used to construct the correlation between causal features of multiple samples; B, l represents the mini-batch of the training dataset. con This refers to consistency constraints.
[0069] The target loss function is expressed as follows:
[0070]
[0071] Wherein, L refers to the target loss function.
[0072] Step S105: Based on the target loss function, update the network parameters in the initial student model, and use the updated initial student model as the trained student model.
[0073] In one embodiment, after the step of updating the network parameters in the initial student model based on the target loss function and using the updated initial student model as the trained student model, the method includes:
[0074] Step S201: Acquire medical images;
[0075] Step S202: Based on the medical image and the trained student model, make a prediction to obtain the target prediction distribution corresponding to the third causal feature.
[0076] In one embodiment, the proposed student model has been experimentally verified to be feasible. It achieved state-of-the-art semi-supervised learning performance on three medical image classification datasets. On the ISIC dataset, the proposed causal-driven consistency method achieved optimal classification performance under various labeled sample configurations (5%, 10%, 20%, 30%). Specifically, in all labeled data ratio settings, the student model of this invention improved the F1 score by almost 3% compared to the baseline. Compared to other state-of-the-art semi-supervised methods such as SimMatch, ACPL, FixMatch, and TEAR, the framework of this invention consistently achieved superior results. With labeled data settings of 5% and 30%, the student model of this invention outperformed FixMatch and ACPL by 2.33%-2.87% and 2.09%-3.35% in F1 scores, respectively. Compared to TEAR, the student model of this invention outperformed it by 1.09%-1.74% and 1.47%-3.13% with labeled data settings of 10% and 20%, respectively. Consistent with the model and settings used in ISIC, experiments were conducted on the Kvasir-Capsule dataset. With a minimum of 5% labeled data, the student model of this invention achieved an improvement of 3.56%–5.14% in F1 score compared to the baseline model. When the available labeled samples reached 30%, the student model of this invention showed a more significant improvement (4.11%–7.68%) compared to the original model.
[0077] It should be noted that this invention overcomes the shortcomings of traditional deep learning models based on statistical dependency generation in establishing spurious relationships between learned features and labels. The invented method focuses on semantically invariant causal features in images through causal learning, improving the model's generalization ability and interpretability. This invention proposes a novel causal relationship-driven consistency regularization semi-supervised learning paradigm. This innovative method can be universally applied to existing consistency regularization-based semi-supervised architectures, improving their efficiency and accuracy. This invention enhances the effect of causal learning by integrating decoupled representations of mutual information constraints during the supervision process. Simultaneously, to effectively utilize the semantic information in unlabeled samples, two general causal-driven consistency mechanisms are developed.
[0078] It should also be noted that traditional semi-supervised models based on consistency regularization do not explicitly distinguish between different causal and non-causal features. Therefore, those non-causal but distinguishable features may replace inherent causal semantic information through invariant constraints, leading to spurious consistency alignment learning. Thus, when utilizing unlabeled data, applying traditional consistency constraints to alignment-enhanced sample features or predicted distributions may result in suboptimal learning performance. To address these issues, this invention proposes a novel causal-driven semi-supervised learning framework to improve the consistency learning process and reduce the impact of confounding factors or biases on effective medical image classification performance. Combining causal inference, the framework emphasizes the extraction and utilization of causal features, thereby improving the model's generalization ability and interpretability. The proposed causal-driven consistency paradigm can be independently designed or applied to existing consistency regularization-based semi-supervised architectures to improve their efficiency and accuracy without significantly increasing parameters and computational burden.
[0079] Please see Figure 4 As shown, in one embodiment, a semi-supervised image analysis device is provided, the device comprising: an acquisition module 10, used to acquire sample images in a training set;
[0080] The first prediction module 20 is used to make predictions based on the sample image and the initial student model to obtain a first prediction distribution corresponding to the non-causal feature and a second prediction distribution corresponding to the first causal feature.
[0081] The second prediction module 30 is used to make predictions based on the sample image and the initial teacher model to obtain the third prediction distribution corresponding to the second causal feature.
[0082] The determination module 40 is used to determine the target loss function based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature;
[0083] The training module 50 is used to update the network parameters in the initial student model based on the target loss function, and use the updated initial student model as the trained student model.
[0084] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a semi-supervised image analysis method on the server side.
[0085] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements client-side functions or steps of a semi-supervised image analysis method.
[0086] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the following steps:
[0087] Obtain sample images from the training set;
[0088] Based on the sample images and the initial student model, predictions are made to obtain the first prediction distribution corresponding to the non-causal features and the second prediction distribution corresponding to the first causal features.
[0089] Based on the sample images and the initial teacher model, a third prediction distribution corresponding to the second causal feature is obtained;
[0090] Based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature, the target loss function is determined;
[0091] Based on the target loss function, the network parameters in the initial student model are updated, and the updated initial student model is used as the trained student model.
[0092] In one embodiment, a computer-readable storage medium is provided that stores a computer program, which, when executed by a processor, performs the following steps:
[0093] Obtain sample images from the training set;
[0094] Based on the sample images and the initial student model, predictions are made to obtain the first prediction distribution corresponding to the non-causal features and the second prediction distribution corresponding to the first causal features.
[0095] Based on the sample images and the initial teacher model, a third prediction distribution corresponding to the second causal feature is obtained;
[0096] Based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature, the target loss function is determined;
[0097] Based on the target loss function, the network parameters in the initial student model are updated, and the updated initial student model is used as the trained student model.
[0098] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.
[0099] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0100] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0101] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.
Claims
1. A semi-supervised image analysis method, characterized in that, The semi-supervised image analysis method includes: Obtain sample images from the training set; Based on the sample images and the initial student model, predictions are made to obtain the first prediction distribution corresponding to the non-causal features and the second prediction distribution corresponding to the first causal features. Based on the sample images and the initial teacher model, a third prediction distribution corresponding to the second causal feature is obtained; Based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature, the target loss function is determined; Based on the target loss function, the network parameters in the initial student model are updated, and the updated initial student model is used as the trained student model.
2. The semi-supervised image analysis method according to claim 1, characterized in that, The step of making predictions based on the sample images and the initial student model to obtain the first prediction distribution corresponding to the non-causal features and the second prediction distribution corresponding to the first causal features includes: Based on the sample image input and the encoder in the initial student model, the first feature is obtained; Based on the first feature and the scoring function, a first mask is obtained, and based on the first mask, a second mask is determined; Based on the first feature and the first mask, a first causal feature is obtained, and based on the first feature and the second mask, a non-causal feature is obtained. Based on the first linear classifier and non-causal features in the initial student model, the first prediction distribution corresponding to the non-causal features is obtained; Based on the second linear classifier and the first causal feature in the initial student model, the second prediction distribution corresponding to the first causal feature is obtained.
3. The semi-supervised image analysis method according to claim 2, characterized in that, The step of predicting the third prediction distribution corresponding to the second causal feature based on the sample image and the initial teacher model: Based on the sample image input and the encoder in the initial teacher model, the second feature is obtained; Based on the second feature and the first mask obtained by sharing the initial teacher model, the second causal feature is obtained; Based on the third linear classifier and the second causal feature in the initial teacher model, the third prediction distribution corresponding to the second causal feature is obtained.
4. The semi-supervised image analysis method according to claim 3, characterized in that, The objective loss function includes a supervised loss, which is expressed as follows: in, Represented as the second prediction distribution, Represents cross-entropy loss, Indicates the true label, This indicates the number of labeled samples. For sample images, To monitor losses; The target loss function also includes an alignment loss, which is expressed as follows: in, The first prediction distribution, To ensure uniform distribution, Represents the Kullback-Leibler entropy. For alignment loss.
5. The semi-supervised image analysis method according to claim 4, characterized in that, The target loss function further includes a mutual information function, which is expressed as follows: in, A variational approximation representing the true distribution. This refers to the first causal characteristic. This refers to non-causal characteristics. It refers to mutual information. This refers to the mutual information function.
6. The semi-supervised image analysis method according to claim 5, characterized in that, The objective loss function further includes a first causal consistency alignment loss, which is expressed as follows: in, The number of unlabeled samples in the training set. This refers to the third prediction distribution. This represents the first causal consistency alignment loss; The objective loss function also includes a second causal consistency alignment loss, which is expressed as follows: in, The second causal consistency alignment loss describes the inter-sample consistency mechanism based on causal alignment, which constrains the consistency of causal feature outputs from different branches. R Represents a relation definition function used to construct the correlation between causal features of multiple samples; mini-batch training datasets. B , This refers to consistency constraints; The target loss function is expressed as follows: Among them, the This refers to the target loss function.
7. The semi-supervised image analysis method according to any one of claims 1 to 6, characterized in that, Following the step of updating the network parameters in the initial student model based on the target loss function and using the updated initial student model as the trained student model, the following steps are included: Acquiring medical images; Based on the medical images and the trained student model, a target prediction distribution corresponding to the third causal feature is obtained.
8. A semi-supervised image analysis device, characterized in that, The semi-supervised image analysis device includes: The acquisition module is used to acquire sample images from the training set; The first prediction module is used to make predictions based on the sample images and the initial student model to obtain a first prediction distribution corresponding to non-causal features and a second prediction distribution corresponding to the first causal features. The second prediction module is used to make predictions based on the sample images and the initial teacher model to obtain the third prediction distribution corresponding to the second causal feature. The determination module is used to determine the target loss function based on the first prediction distribution, the second prediction distribution, the third prediction distribution, the first causal feature, the second causal feature, and the non-causal feature; The training module is used to update the network parameters in the initial student model based on the target loss function, and use the updated initial student model as the trained student model.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the semi-supervised image analysis method as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the semi-supervised image analysis method as described in any one of claims 1 to 7.